Raid 10 vs SHR(raid5) on Synology NAS for VM Disks

[potetpro]

Hi.



For info we have 2x synology rackstation rs2421rp+, both with 8x 8TB Ironwolf disks.
The NAS are connected via 10gbit to the proxmox servers.

These are gonna be put in Synology HA mode. But before that we are testing each of them in different RAID setup to check performance.
The share is gonna be used for VM disks, both storage for customers, and for customers not willing to pay for SSD OS disk.


So i guess many of you are gonna say Raid10, because of rebuild time, and performance when scaling the system.
But then i found this thread:
https://www.reddit.com/r/synology/comments/i70bow/maximum_nas_performance_shr_vs_raid10/

And when testing for myself as well, SHR seem to perform close to Raid10, in some cases:

1GB file:



16GB file:


When trying to copy system32 from windows to C:\ they finished almost at the same time, 3:54 vs 4:15,

 

[ness1602]

If i'm not mistaken, SHR is not Raid5, more like JBOD/snapraid with some parity functionality.

 

[potetpro]

It's just RAID5 + LVM

 

[potetpro]

 

[ness1602]

Usually Raid10 is used for performance, and raid5 for space usability. The rebuild time for 8tb Ironwolf in zfs is around 16hours if i'm not mistaken, i've done it this year. So you need to estimate if this is okay for you. If you insist on raid5(i use it in a lot PBS and OMV deployments, not that much for VM/CT), i would recommend 7disks raid5 and one hot spare,just in case.

 

[apoc]

Your tests do not reflect any real life scenario nor do they show a significant difference, due to the tool used and the limit time of testing / working set size.

Crystaldiskmark is a consumer utility in my eyes and the results mean nothing.

Usually Raid10 is used for performance, and raid5 for space usability

I disagree. Raid 5 also can be used for performance - if the workload is the right one it is not less performant. So it is all about knowing and understanding the workload.

If you have an idea about it, you can do some proper testing and get some values which will bring you forward.
Everything else is just messing around creating some numbers

 

[UdoB]

The share is gonna be used for VM disks

In this case you should NOT look at "Gigabytes per Second" but IOPS! A Raid10 with 4*2 disks will probably have ~ four times better performance than a Raid5/SHR. This is just because in Raid5 those disks are not independent, but in Raid10 every mirror (not every device) can do its job in parallel.
You may perhaps compensate this partially with a mirrored SSD as special device inside the Synology. (I have no experience with this, so I say "perhaps".)
(And be careful interpreting Crystal as it often tests your Ram and not really your disk performance...)

Just my two €¢...

 

[potetpro]

In this case you should NOT look at "Gigabytes per Second" but IOPS! A Raid10 with 4*2 disks will probably have ~ four times better performance than a Raid5/SHR. This is just because in Raid5 those disks are not independent, but in Raid10 every mirror (not every device) can do its job in parallel.
You may perhaps compensate this partially with a mirrored SSD as special device inside the Synology. (I have no experience with this, so I say "perhaps".)
(And be careful interpreting Crystal as it often tests your Ram and not really your disk performance...)

Just my two €¢...

Regarding the IOPS.
The Raid10 is faster, now testing with 32GB to to way beyond the cache size.

SHR:


Raid10:


Disregard the 0.00, accedetly press the button to run this again, and canceled.
It was close to the value of the SHR.